|
Here's a fun dall-e thing: 1) Generate a bunch of Isometric Orthographic images. You can do this with prompts formatted thusly: "Exterior othographic isometric scene. [whatever content & style]" 2) Scale them down to preferred sizes and arrange them around a 16x16pixel grid with transparency in between (just make sure they align as closely as possible with the 16x16 grid) (I don't mean a 16x16 grid of chunks, I mean a grid of 16x16 pixel chunks, in case that's not clear - dall-e essentially creates a mural of 16x16 pixel chunks so you want your source images to align with that mural's alignment as closely as possible) 3) Upload to Dall-E and repeat the prompt but prompt for a larger scale ("a kingdom" instead of "a castle") deep dish peat moss has a new favorite as of 21:07 on Aug 24, 2022 |
# ? Aug 24, 2022 19:36 |
|
|
# ? May 28, 2024 08:29 |
|
|
# ? Aug 24, 2022 19:54 |
|
deep dish peat moss posted:Here's a fun dall-e thing: whoa, that's a really cool use case
|
# ? Aug 24, 2022 19:55 |
|
Here's a better example of it since I need to write it up for a patreon post anyway First I generated 3 isometric scenes prompts which gave me these images I used in the final piece: Exterior othographic isometric scene of 2090s neo-amsterdam with crowds of pedestrians. :: Style: Akira Toriyama, SciFi, Dragonball, Digital Art Exterior othographic isometric scene of 1980s tokyo with crowds of pedestrians. :: Style: Akira Toriyama, SciFi, Dragonball, Digital Art Exterior othographic isometric view. Megacity slum tenement. :: Style: Shadowrun, Cyberpunk, Scifi, Futuristic, Digital Art I cropped the bits I wanted to keep out of them, then scaled them down so they were roughly even in size. Then I aligned them along a 16x16 pixel grid: I uploaded that to dall-e and tried a few different prompts: Exterior othographic isometric map. A futuristic city with traditional architecture styles. :: Style: Akira Toriyama, SciFi, Dragonball, Digital Art Exterior othographic isometric city map. A futuristic city with traditional architecture styles. :: Style: Akira Toriyama, SciFi, Dragonball, Digital Art. Toonart. Exterior othographic isometric district map. A futuristic city with traditional architecture styles. :: Style: Akira Toriyama, SciFi, Dragonball, Digital Art. Toonart. (e: and here it is on patreon where all my posts about stuff like this are free ) deep dish peat moss has a new favorite as of 22:29 on Aug 24, 2022 |
# ? Aug 24, 2022 21:52 |
|
Am I weird or is the common prompt lexicography becoming more interesting than the literal outputs? We're basically at the point you can get computer to show you Tayne with hat wobble if you speak the language.
|
# ? Aug 24, 2022 21:59 |
|
That is all super awesome, one of the neater things I've seen with all these art generators.
|
# ? Aug 24, 2022 22:00 |
|
zedprime posted:Am I weird or is the common prompt lexicography becoming more interesting than the literal outputs? There are people literally using emojis as a sort of math operator for prompt words and it's blowing my mind I poked at the idea a little but the way dall-e parses emojis is bizarre and alien to me. But yeah, same - I'm not too interested in the actual outputs themselves (beyond "they're neat!") but I am loving the weird word-logic-math-programming puzzle involved in figuring out how to get there, as well as putting together use cases for how AI art can be used beyond just generating "neat images" I am unbelievably disappointed that so much of the usage I see in the DALL-E discord is just things like people making fanart of their favorite anime characters or putting their favorite anime characters in different outfits or creating their CUSTOM ORIGINAL CHARACTER DO NOT STEAL fanfic characters that are extremely derivative of their favorite anime because they literally just typed "[characterdescription] from [theirfavoriteanime]" - and those posts seem to be very highly rated too I'm also pretty bored of all the people just doing "[Description], digital art" because it doesn't produce repeatable styles and it's the exact kind of thing that makes the actual art community distrustful of AI art - they see it as a way of stealing other artists' work instead of as a tool for creating something new and unique. The DALL-E subreddit is pretty much exclusively dall-e made meme images. All the posts on my Patreon are about pretty much about either how DALL-E interprets syntax, symbology, formatting, etc., or a specific focus on doing it all without being distinctly derivative of any existing artist's artstyle, and they're all free, but I am too broke to buy credits much longer unless I find a way to monetize this stuff and it has for the most part turned out to be too highbrow or esoteric for the DALL-E Discord/Reddit crowds https://www.patreon.com/benh1987 When I posted my initial experiments with this isometric map concept last night I had 2 different DMs on Discord from people telling me what a cool guide it was and how much it helped them, and they sent the images they made with it and it was literally just... placing two screenshots from two different animes in separate corners of the canvas and letting dall-e fill in the middle Which is one of the basic editing techniques discussed in the official dall-e prompt book and has very little to do with creating isometric cities and maps, a technique that I showcased specifically because of the orthographic perspective's ability for objects to be moved around the scene without worrying about shifting perspective. deep dish peat moss has a new favorite as of 03:27 on Aug 25, 2022 |
# ? Aug 24, 2022 22:34 |
|
deep dish peat moss posted:I am unbelievably disappointed that so much of the usage I see in the DALL-E discord is just things like people making fanart of their favorite anime characters or putting their favorite anime characters in different outfits or creating their CUSTOM ORIGINAL CHARACTER DO NOT STEAL fanfic characters that are extremely derivative of their favorite anime because they literally just typed "[characterdescription] from [theirfavoriteanime]" - and those posts seem to be very highly rated too I'm also pretty bored of all the people just doing "[Description], digital art" because it doesn't produce repeatable styles and it's the exact kind of thing that makes the actual art community distrustful of AI art - they see it as a way of stealing other artists' work instead of as a tool for creating something new and unique. The DALL-E subreddit is pretty much exclusively dall-e made meme images. Literally monkeys at keyboards.
|
# ? Aug 24, 2022 23:15 |
|
Kinda liking that prompt, they're doin' their best!
|
# ? Aug 24, 2022 23:52 |
|
Did the AI attempt a loss edit of the word "loss"?
|
# ? Aug 25, 2022 00:09 |
|
Wasabi the J posted:Did the AI attempt a loss edit of the word "loss"? All four images were the result of Is this loss?, I tried to arrange them in the closest order I could
|
# ? Aug 25, 2022 00:41 |
|
Now that pretty much everything on MidJourney's community feed is just absolutely stunning, I'm noticing that a lot of the parent images are simply the same image. I used to enjoy looking through the process the user took to get there, and that seems to be missing on most of the community feed now. I've read through the documentation and a couple guides on developing good images, but I feel like anytime one of my images gets anywhere near that quality it was just by accident. If anybody could point me toward an explanation, I'd appreciate it. I'm only buying a month right now and then am going to spend some time with my creations turning some of them into "real" art, and I only have a week left. I'm just feel like I've missed some critical step in getting these amazing pieces like on the feed. Thanks in advance.
|
# ? Aug 25, 2022 01:32 |
|
deep dish peat moss posted:I am unbelievably disappointed that so much of the usage I see in the DALL-E discord is just things like people making fanart of their favorite anime characters or putting their favorite anime characters in different outfits or creating their CUSTOM ORIGINAL CHARACTER DO NOT STEAL fanfic characters that are extremely derivative of their favorite anime because they literally just typed "[characterdescription] from [theirfavoriteanime]" - and those posts seem to be very highly rated too I'm also pretty bored of all the people just doing "[Description], digital art" because it doesn't produce repeatable styles and it's the exact kind of thing that makes the actual art community distrustful of AI art - they see it as a way of stealing other artists' work instead of as a tool for creating something new and unique. The DALL-E subreddit is pretty much exclusively dall-e made meme images.
|
# ? Aug 25, 2022 05:28 |
|
It's also fine to be disappointed by the fact that none of them are actually engaging with what the toy does Anyway SD is actually pretty dang good for some things, particularly traditionally-illustrated styles which dall-e can't do well at all
|
# ? Aug 25, 2022 06:17 |
|
deep dish peat moss posted:It's also fine to be disappointed by the fact that none of them are actually engaging with what the toy does can you share some of these prompts please
|
# ? Aug 25, 2022 07:11 |
|
Someone sneaking out of their suburban home under the watchful gaze of an eldritchgod. Inspired by lovecraft, h.r. giger, and Stephen Gammell A spacemarine cautiously approaching a bizarre meaty monolith on the lunar surface. Inspired by lovecraft, h.r. giger and scary stories to tell in the dark View down a corridor of a derelict space hulk infested with eldritch lovecraftian beasts. (these were technically made with https://beta.dreamstudio.ai/ - I went through all my free credits with it, it's neat, if I had money to play with I'd throw some at it right now ) e: Oh, you can just sign up with another email for another £2 in free credits. Heads up that generating multiple images takes 1 credit per image e2: After a bunch more playing around, I really like SD. It doesn't have the flexibility dall-e has but it can handle fine detail and therefore some styles like anything pencil/pen related waaaaay better. It's nice being able to generate individual images at 10steps for 0.2 credits each then doing a full 1-credit 50 step generation of the same seed when you get one you like. That being said, "pencil art of bizarre Giger-esque astronauts fighting eldritch horrors on the moon and mars" is the only consistently cool style I've found for it. And everything it does regardless of prompt is extremely desaturated. prompts: call the astronauts "space marines" insert the word Eldritch a lot Insert whatever macabre or lovecraftian words you want Describe things as gross or fleshy or hideous or horrifying or bizarre or meaty or alien or incomprehensible, etc End with "Inspired by lovecraft, photorealistic pencil art, h.r. giger, john berkey, and scary stories to tell in the dark" Results of the same prompt that generated a lot of the above used on dall-e: deep dish peat moss has a new favorite as of 13:56 on Aug 25, 2022 |
# ? Aug 25, 2022 08:44 |
|
"Deliriously happy woman, open mouth smile."
|
# ? Aug 25, 2022 23:50 |
|
Dick Trauma posted:"Deliriously happy woman, open mouth smile." I always suspected women were xenomorphs
|
# ? Aug 26, 2022 00:19 |
|
ikanreed posted:I always suspected women were xenomorphs [Ripley pulls back from the door, readying her rifle before peering back inside the chamber. Salads? The Xenos are eating salads and... laughing!]
|
# ? Aug 26, 2022 02:09 |
|
|
# ? Aug 26, 2022 02:42 |
|
Digging Stable Diffusion. Maybe I'll try the local version some time. A couple older ones:
|
# ? Aug 26, 2022 08:58 |
|
|
# ? Aug 26, 2022 22:06 |
|
Captain Jean-Luc Picard: Commander William Riker: Lieutenant Commander Data: Lieutenant Commander Geordi LaForge: Worf, the Klingon: Doctor Beverly Crusher: Counselor Deanna Troi: Guinan: A Borg: A Ferengi: A Romulan:
|
# ? Aug 27, 2022 02:42 |
|
Just stop trying to make metaverse a thing already
|
# ? Aug 27, 2022 02:45 |
|
StoryTime posted:I fed all the rapsheets on the site into a neural net based moderator AI, let's check out the results.
|
# ? Aug 27, 2022 19:56 |
|
ok I am pretty confident by now that I like StableDiffusion a lot more than dall-e when it comes to generating usable art. Dall-E generations are still wild and imaginative and cool as hell but as far as illustration styles go they function best as concept art. My primary concern with SD is that it's more derivative of artist influences than Dall-E is. Here's some SD pictures of me visualizing a scifi setting I started writing a story in a long time ago and abandoned. tl;dr - dying earth setting where the surface of the earth was melted into sheets of salt and saline bogs and mutant cactus jungles and coral reefs (because all the oceans are dried up duh). And they're amazing but ultimately I feel they're a little derivative of the artists I tagged in the prompt (john berkey, roger dean, and akira toriyama, the first two in particular) Of course SD images are much smaller (even at max size, which gets expensive to generate) so I guess I need to look into a good AI Upscaler next deep dish peat moss has a new favorite as of 21:25 on Aug 27, 2022 |
# ? Aug 27, 2022 21:18 |
|
have you tried midjourney? it's much closer to SD for art like that than dall-e is
|
# ? Aug 27, 2022 22:03 |
|
https://twitter.com/jsmooth995/status/1563351458170318848
|
# ? Aug 28, 2022 00:40 |
|
drone-assisted art https://twitter.com/RustyRuu/status/1563256882134888449
|
# ? Aug 28, 2022 00:52 |
|
Can you guess the song lyrics I used to generate these images using Stable Diffusion?
|
# ? Aug 28, 2022 01:33 |
|
|
# ? Aug 28, 2022 10:18 |
|
https://twitter.com/bgavran3/status/1563640767188451328
|
# ? Aug 28, 2022 14:12 |
|
Brawnfire has a new favorite as of 21:46 on Aug 28, 2022 |
# ? Aug 28, 2022 21:42 |
|
Brawnfire posted:Lieutenant Commander Geordi LaForge: https://www.youtube.com/watch?v=73UfgMoWv8E
|
# ? Aug 28, 2022 21:46 |
|
Not strictly generative but I keep revisiting the state of the art in upscalers every couple of month and the temporal stability has gotten a lot better since even a year ago. Its still not perfect but I was able to turn this: https://i.imgur.com/xgg667R.mp4 Into this: https://i.imgur.com/ZA1I7kC.mp4 Which is quite impressive I think. I think this might finally be good enough to start restoring some old lost media to something reasonably close to the original source.
|
# ? Aug 29, 2022 01:37 |
|
So-Bin the artist who does the book covers for the Overlord book series works really well in a Midjourney prompt. Beholder, So-Bin, Overlord Wizard, So-Bin, Overlord Arachne, So-Bin, Overlord Barbarian, So-Bin, Overlord Alien, So-Bin, Overlord Mind-Flayer, So-Bin, Overlord
|
# ? Aug 29, 2022 14:00 |
|
Just playing Starcraft
|
# ? Aug 29, 2022 14:50 |
|
A spontaneous SD generated meme that made me laugh:
|
# ? Aug 29, 2022 17:40 |
|
|
# ? Aug 29, 2022 19:19 |
|
|
# ? May 28, 2024 08:29 |
|
quote:i'll have you know i graduated top of my class in the navy seals, and i've been involved in numerous secret raids on al-quaeda, and i have over 300 confirmed kills. i am trained in gorilla warfare and i'm the top sniper
|
# ? Aug 29, 2022 20:19 |